A Study Of Data Pattern Analysis On Urban Road Traffic Status

Urban road networks are transportation infrastructure to ensure kinds of activities in daily life. In China, traffic demand amount has dramatically increased during the course of Chinese urbanization and space expansion, as a consequence, which has aroused obvious traffic conflict and knottiness except for the restriction of urban sustainable development in major metropolitan areas. For the sake of improving the traffic capability, intelligent transportation system (ITS) has been proposed. It is the application that incorporates electronic, computer, and communication technologies into vehicles and roadways for monitoring traffic conditions, reducing congestion, enhancing mobility, and so on. To achieve these goals, in past decades, there have been many approaches proposed for tackling related problems in ITS, and today it is internationally recognized as the best way to solve these traffic problems. Along with the construction of urban transportation informationization, traffic detector types are various, and they consist of fixed and mobile detetors in which floacting car is reletively advanced instrument to gather roadway information in ITS. Due to the continuous progress in informationization communities, diverse traffic data acquirision modes make it possible provide reliable information sources for transportation departments and decision-makers, so it is inevitable to explore core technologies and methods for vast and multi-source traffic data.Faced with those above problems and requirement, this dissertation proposes the research topic on traffic status pattern analysis in urban road network. In transportation system, one of the most difficult tasks is to handle the mass amount of growing data and discover useful information from them. Data fusion technology is a kind of information comprehensive processing method, which can be used to deal with multiple sources data from traffic sensors for improving data quality and boosting up information efficiency in use. Data mining is the exploration and analysis of large quantities of data in order to discover valid, novel, potentially useful, and ultimately understandable patterns in data. At present data mining has been already a hot research means in analzing traffic data. Traffic flow is a complex, changeful, nonlinear, unstructured, space time-varying and random system. The evolution of traffic status follows certain temporal-spatial rules and patterns, and the challenge is to extract such patterns by processing and analzing mass traffic data. In this view, it is significant to practically analyze traffic patterns for ultimate traffic management and services by combining data fusion, data mining and GIS technologies. Therefore, this dissertation makes a study of three key issues including data fusion, pattern classification and spatial patter distribution of traffic status on urban road networks. Main research work in the dissertation is as follows:(1) Technically, ITS progress and application role in the world are introduced. In the view of transportation background in Chinese cities, three key issues including data fusion, pattern classification and spatial distribution of traffic status on urban road networks are proposed after problems and trend of informationization construction in transportation are elabrated. Then research progress related to those key issues is detailedly reviewed, on which research objectives and contents of the dissertation are put forward.(2) The conceptual describtion of traffic status in different harartical level are defined. Three important characteristic of traffic status and main traffic flow parameters are summarized. It studies present fixed and mobile detectors in transportation system and analzes their advantages and shortcomings. Based on traffic data type depiction, it designs a framework model for traffic status pattern analysis, meanwhile, and introduces basic data processing and analysis technologies in the framework.(3) Based on comparing the variety of transportation detectors and generalizing key technologies of GPS floating car, it concludes the actual requirement of integrating multi-source traffic data merits, especially for fixed and mobile detectors. It proposes a two-dimension traffic data fusion model based on support vector regression algorithm that has a good nonliear modeling and can avoid local optimal problem after reviewing data fusion theory and existing methods. Supported by microcosmic traffic simulation software PTV VISSIM, it designs an expreriment schema and acquires simulating traffic varable data of loop detector and floating car. The processed data are tested to perform experiments on sampling time interval, road segments and fusion algorithms and the estimation errors are compared.(4) It analyzes evaluation research on road traffic conditions, and designs an evaluation frame of urban traffic status levels with velocity range. It presents uncertainty problem in traffic status linguistic expression, and adopts fuzy sets strategy to represent traffic measurable factor with membership function. With regard to the training limitation of fuzzy classification, it proposes a multi-class classification model for traffic status with fuzzy support vector machine. The simulation average and traveling speed data are tested to identify multi-class label and classification accuracy with different algorithms are discussed.(5) It describes spatial distribution of geographic objects and introduces traffic status points’ distribution on urban road network. Based on conventional spatial density clustering algorithms, it puts forward a road-constrained DBSCAN (R-DBSCAN) algorithm to adapt spatial distribution computing, in which graph theory is used to model spatial geometric constrain and traffic behavior constrain with network topology. R-DBSCAN algorithm measures spatal point similarity with shortest network distance, which deals with geospatial influence relationship between traffic status and road network. The experiment takes real data collected from GPS floating cars in Nanchang city as a case study, datapoints distribution properties on traffic congestion status are analyzed in comarision with R-DBSCAN algorithm imlemented in GIS envrionment.